arXiv subjects

Yi Xie

Publications and source records attributed to Yi Xie.

At least 55 records · Page 3Linked to original sources

arXiv2026

LifeBench: A Benchmark for Long-Horizon Multi-Source Memory

Long-term memory is fundamental for personalized agents capable of accumulating knowledge, reasoning over user experiences, and adapting across time. However, existing memory benchmarks primarily target declarative memory, specifically semantic and episodic types, where all information is explicitly presented in dialogues. In contrast, real-world actions are also governed by non-declarative memory, including habitual and procedural types, and need to be inferred from diverse digital traces. To bridge this gap, we introduce Lifebench, which features densely connected, long-horizon event simulation. It pushes AI agents beyond simple recall, requiring the integration of declarative and non-declarative memory reasoning across diverse and temporally extended contexts. Building such a benchmark presents two key challenges: ensuring data quality and scalability. We maintain data quality by employing real-world priors, including anonymized social surveys, map APIs, and holiday-integrated calendars, thus enforcing fidelity, diversity and behavioral rationality within the dataset. Towards scalability, we draw inspiration from cognitive science and structure events according to their partonomic hierarchy; enabling efficient parallel generation while maintaining global coherence. Performance results show that top-tier, state-of-the-art memory systems reach just 55.2\% accuracy, highlighting the inherent difficulty of long-horizon retrieval and multi-source integration within our proposed benchmark. The dataset and data synthesis code are available at https://github.com/1754955896/LifeBench.

cs.AI↗

arXiv2026

REPAIR: Robust Editing via Progressive Adaptive Intervention and Reintegration

Post-training for large language models (LLMs) is constrained by the high cost of acquiring new knowledge or correcting errors and by the unintended side effects that frequently arise from retraining. To address these issues, we introduce REPAIR (Robust Editing via Progressive Adaptive Intervention and Reintegration), a lifelong editing framework designed to support precise and low-cost model updates while preserving non-target knowledge. REPAIR mitigates the instability and conflicts of large-scale sequential edits through a closed-loop feedback mechanism coupled with dynamic memory management. Furthermore, by incorporating frequent knowledge fusion and enforcing strong locality guards, REPAIR effectively addresses the shortcomings of traditional distribution-agnostic approaches that often overlook unintended ripple effects. Our experiments demonstrate that REPAIR boosts editing accuracy by 10%-30% across multiple model families and significantly reduces knowledge forgetting. This work introduces a robust framework for developing reliable, scalable, and continually evolving LLMs.

cs.CL↗

arXiv2026

Pseudo-isotopies of 3-manifolds with infinite fundamental groups

Suppose $Y$ is a compact, connected, oriented 3-manifold possibly with boundary, such that $π_1(Y)$ is infinite. Let $\operatorname{Diff}_\partial(I\times Y)$ denote the group of self-diffeomorphisms of $I\times Y$ that are equal to the identity near the boundary. Let $\operatorname{Diff}_{PI}(I\times Y)$ denote the subgroup of $\operatorname{Diff}_\partial(I\times Y)$ consisting of elements pseudo-isotopic to the identity. Define $\operatorname{Homeo}_\partial(I\times Y)$, $\operatorname{Homeo}_{PI}(I\times Y)$ similarly for homeomorphisms. We show that the canonical map $π_0\operatorname{Diff}_{PI}(I\times Y) \to π_0\operatorname{Homeo}_{PI}(I\times Y)$ is of infinite rank. As a consequence, $π_0\operatorname{Diff}_{PI}(I\times Y)$, $π_0\operatorname{Diff}_{\partial}(I\times Y)$, $π_0\operatorname{Homeo}_{PI}(I\times Y)$, $π_0\operatorname{Homeo}_{\partial}(I\times Y)$ are all abelian groups of infinite rank. We also prove that $π_0\,C(Y)$ contains an abelian subgroup of infinite rank, and $π_0\,C(I\times Y)$ admits a surjection to an abelian group of infinite rank, where $C(X)$ denotes the concordance automorphism group $\operatorname{Diff}(I\times X, \{0\}\times X\cup I\times \partial X)$ or $\operatorname{Homeo}(I\times X, \{0\}\times X\cup I\times \partial X)$. These results are proved by studying the actions of barbell diffeomorphisms on the spaces of embedded arcs and configuration spaces.

math.GT↗

arXiv2026

ShapeCond: Fast Shapelet-Guided Dataset Condensation for Time Series Classification

Time series data supports many domains (e.g., finance and climate science), but its rapid growth strains storage and computation. Dataset condensation can alleviate this by synthesizing a compact training set that preserves key information. Yet most condensation methods are image-centric and often fail on time series because they miss time-series-specific temporal structure, especially local discriminative motifs such as shapelets. In this work, we propose ShapeCond, a novel and efficient condensation framework for time series classification that leverages shapelet-based dataset knowledge via a shapelet-guided optimization strategy. Our shapelet-assisted synthesis cost is independent of sequence length: longer series yield larger speedups in synthesis (e.g., 29$\times$ faster over prior state-of-the-art method CondTSC for time-series condensation, and up to 10,000$\times$ over naively using shapelets on the Sleep dataset with 3,000 timesteps). By explicitly preserving critical local patterns, ShapeCond improves downstream accuracy and consistently outperforms all prior state-of-the-art time series dataset condensation methods across extensive experiments. Code is available at https://github.com/lunaaa95/ShapeCond.

cs.LG↗

arXiv2026

Dax invariants, light bulbs, and isotopies of symplectic structures

This paper addresses several isotopy problems on $4$-manifolds. First, we classify the isotopy classes of embeddings of $Σ$ in $Σ\times S^2$ that are geometrically dual to $\{\mbox{pt}\}\times S^2$, where $Σ$ is a closed oriented surface with a positive genus, and show that there exist infinitely many such embeddings that are homotopic to each other but mutually non-isotopic, thereby answering a question of Gabai. By combining this construction with techniques from symplectic topology, we also answer Problem 2(a) in McDuff-Salamon's problem list and a question of Cieliebak-Eliashberg-Mishachev, which concern the uniqueness and $h$-principle of symplectic structures on closed $4$-manifolds. We answer these questions by establishing the following results: (1) The space of symplectic forms on every irrational ruled surface homologous to a fixed symplectic form has infinitely many connected components; (2) There exist infinitely many symplectic forms on every irrational ruled surface that are formally homotopic, cohomologous, but not homotopic to each other. Both are the first such examples for closed $4$-manifolds. The proofs are based on a generalization of the Dax invariant to embedded closed surfaces. In the course of the proof, we obtain several properties of the smooth mapping class group of $Σ\times S^2$, which may be of independent interest. For example, we show that there exists a surjective homomorphism from $π_0\operatorname{Diff}(Σ\times S^2)$ to $\mathbb{Z}^\infty$, such that its restriction to the subgroup of elements pseudo-isotopic to the identity is of infinite rank.

math.GT↗

arXiv2026

WADBERT: Dual-channel Web Attack Detection Based on BERT Models

Web attack detection is the first line of defense for securing web applications, designed to preemptively identify malicious activities. Deep learning-based approaches are increasingly popular for their advantages: automatically learning complex patterns and extracting semantic features from HTTP requests to achieve superior detection performance. However, existing methods are less effective in embedding irregular HTTP requests, even failing to model unordered parameters and achieve attack traceability. In this paper, we propose an effective web attack detection model, named WADBERT. It achieves high detection accuracy while enabling the precise identification of malicious parameters. To this end, we first employ Hybrid Granularity Embedding (HGE) to generate fine-grained embeddings for URL and payload parameters. Then, URLBERT and SecBERT are respectively utilized to extract their semantic features. Further, parameter-level features (extracted by SecBERT) are fused through a multi-head attention mechanism, resulting in a comprehensive payload feature. Finally, by feeding the concatenated URL and payload features into a linear classifier, a final detection result is obtained. The experimental results on CSIC2010 and SR-BH2020 datasets validate the efficacy of WADBERT, which respectively achieves F1-scores of 99.63% and 99.50%, and significantly outperforms state-of-the-art methods.

cs.CR↗

arXiv2026

Towards Effective Prompt Stealing Attack against Text-to-Image Diffusion Models

Text-to-Image (T2I) models, represented by DALL$\cdot$E and Midjourney, have gained huge popularity for creating realistic images. The quality of these images relies on the carefully engineered prompts, which have become valuable intellectual property. While skilled prompters showcase their AI-generated art on markets to attract buyers, this business incidentally exposes them to \textit{prompt stealing attacks}. Existing state-of-the-art attack techniques reconstruct the prompts from a fixed set of modifiers (i.e., style descriptions) with model-specific training, which exhibit restricted adaptability and effectiveness to diverse showcases (i.e., target images) and diffusion models. To alleviate these limitations, we propose Prometheus, a training-free, proxy-in-the-loop, search-based prompt-stealing attack, which reverse-engineers the valuable prompts of the showcases by interacting with a local proxy model. It consists of three innovative designs. First, we introduce dynamic modifiers, as a supplement to static modifiers used in prior works. These dynamic modifiers provide more details specific to the showcases, and we exploit NLP analysis to generate them on the fly. Second, we design a contextual matching algorithm to sort both dynamic and static modifiers. This offline process helps reduce the search space of the subsequent step. Third, we interact with a local proxy model to invert the prompts with a greedy search algorithm. Based on the feedback guidance, we refine the prompt to achieve higher fidelity. The evaluation results show that Prometheus successfully extracts prompts from popular platforms like PromptBase and AIFrog against diverse victim models, including Midjourney, Leonardo.ai, and DALL$\cdot$E, with an ASR improvement of 25.0\%. We also validate that Prometheus is resistant to extensive potential defenses, further highlighting its severity in practice.

cs.CR↗

arXiv2026

PERM: Psychology-grounded Empathetic Reward Modeling for Large Language Models

Large Language Models (LLMs) are increasingly deployed in human-centric applications, yet they often fail to provide substantive emotional support. While Reinforcement Learning (RL) has been utilized to enhance empathy of LLMs, existing reward models typically evaluate empathy from a single perspective, overlooking the inherently bidirectional interaction nature of empathy between the supporter and seeker as defined by Empathy Cycle theory. To address this limitation, we propose Psychology-grounded Empathetic Reward Modeling (PERM). PERM operationalizes empathy evaluation through a bidirectional decomposition: 1) Supporter perspective, assessing internal resonation and communicative expression; 2) Seeker perspective, evaluating emotional reception. Additionally, it incorporates a bystander perspective to monitor overall interaction quality. Extensive experiments on a widely-used emotional intelligence benchmark and an industrial daily conversation dataset demonstrate that PERM outperforms state-of-the-art baselines by over 10\%. Furthermore, a blinded user study reveals a 70\% preference for our approach, highlighting its efficacy in generating more empathetic responses. Our code, dataset, and models are available at https://github.com/ZhengWwwq/PERM.

cs.CL↗

arXiv2026

Modality Dominance-Aware Optimization for Embodied RGB-Infrared Perception

RGB-Infrared (RGB-IR) multimodal perception is fundamental to embodied multimedia systems operating in complex physical environments. Although recent cross-modal fusion methods have advanced RGB-IR detection, the optimization dynamics caused by asymmetric modality characteristics remain underexplored. In practice, disparities in information density and feature quality introduce persistent optimization bias, leading training to overemphasize a dominant modality and hindering effective fusion. To quantify this phenomenon, we propose the Modality Dominance Index (MDI), which measures modality dominance by jointly modeling feature entropy and gradient contribution. Based on MDI, we develop a Modality Dominance-Aware Cross-modal Learning (MDACL) framework that regulates cross-modal optimization. MDACL incorporates Hierarchical Cross-modal Guidance (HCG) to enhance feature alignment and Adversarial Equilibrium Regularization (AER) to balance optimization dynamics during fusion. Extensive experiments on three RGB-IR benchmarks demonstrate that MDACL effectively mitigates optimization bias and achieves SOTA performance.

cs.CV↗

arXiv2025

Two birds with one stone: simultaneous realization of both Lunar Coordinate Time and lunar geoid time by a single orbital clock

Context. Among options for definition of the lunar reference time, the option taking Lunar Coordinate Time (O1) has its simplicity but cannot be realized by any clock without steering, while another option adopting the lunar geoid (selenoid) proper time (O2) has its convenience for users on the lunar surface but would bring a new scaling of spatial coordinates and mass parameter of the Moon. Aims. We propose a ''time aligned orbit'' that the readings of an ideal clock in this orbit could equal to the selenoid proper time in O2 and these readings could be converted to Lunar Coordinate Time in O1 by a known linear transformation. Methods. We show that there exist the time aligned orbit around the Moon with its semi-major axis of about 1.5 lunar radius slightly depending on its inclination. We conduct a set of numerical simulations to assess to what extent a clock on these orbits could realize O2 in a more realistic lunar environment. Results. We find that the proper time in our simulations would desynchronize from the selenoid proper time up to 190 ns after a year with a frequency offset of 6E-15, which is solely 3.75% of the frequency difference in O2 caused by the lunar surface topography. These numbers might be further reduced to 13 ns and 4E-16, if we could account for the deviation of the mean orbits in our simulations from the nominal ones. Conclusions. One might simultaneously realize O1 and O2 by deployment of a single clock in the time aligned orbit. This approach also has its scalability for other terrestrial planets beyond the Earth-Moon system.

astro-ph.IM↗

arXiv2025

Spontaneous fission half-lives for heavy and super-heavy nuclei from phenomenological models

A phenomenological model is proposed for a systematic description of the spontaneous fission (SF) half-lives $T_{\rm SF}$ of heavy and super-heavy nuclei. Based on the effective tunneling barrier (ETB), the proposed approach reproduces the SF half-lives of 79 known nuclei with an average deviation of 0.8, which is $17\%$ smaller than that of the linear correlation approach recently proposed in [N. S. Moiseev, N. V. Antonenko and G. G. Adamian, Phys. Rev. C 112, 034607 (2025)]. For superheavy nuclei with $45\leqslant N-Z \leqslant 61$, the predicted SF half-lives from these two different phenomenological models are in good agreement with each other. The ETB calculations implies that the $β$-decay energy affects the SF half-lives of nuclei far from the $β$-stability line. For superheavy nuclei around the magic number $N=184$, the predicted $T_{\rm SF}$ of $^{304}$120 is much shorter than that of $^{298}$Fl. With predicted values of about $10 \sim 160$ ms for $T_{\rm SF}$, the unmeasured SHN $^{293}119 $ could survive for long enough to reach the focal-plane detector in detection systems like the gas-filled recoil separator SHANS in Lanzhou.

nucl-th↗

arXiv2025

Slow Transition to Low-Dimensional Chaos in Heavy-Tailed Recurrent Neural Networks

Growing evidence suggests that synaptic weights in the brain follow heavy-tailed distributions, yet most theoretical analyses of recurrent neural networks (RNNs) assume Gaussian connectivity. We systematically study the activity of RNNs with random weights drawn from biologically plausible Lévy alpha-stable distributions. While mean-field theory for the infinite system predicts that the quiescent state is always unstable -- implying ubiquitous chaos -- our finite-size analysis reveals a sharp transition between quiescent and chaotic dynamics. We theoretically predict the gain at which the system transitions from quiescent to chaotic dynamics, and validate it through simulations. Compared to Gaussian networks, heavy-tailed RNNs exhibit a broader parameter regime near the edge of chaos, namely a slow transition to chaos. However, this robustness comes with a tradeoff: heavier tails reduce the Lyapunov dimension of the attractor, indicating lower effective dimensionality. Our results reveal a biologically aligned tradeoff between the robustness of dynamics near the edge of chaos and the richness of high-dimensional neural activity. By analytically characterizing the transition point in finite-size networks -- where mean-field theory breaks down -- we provide a tractable framework for understanding dynamics in realistically sized, heavy-tailed neural circuits.

q-bio.NC↗

arXiv2025

Lunar Time Ephemeris $\texttt{LTE440}$: definitions, algorithm and performance

Robotic and human activities in the cislunar space are expected to rapidly increase in the future. Modeling, jointly analysis and sharing of time measurements made in the vicinity of the Moon might indispensably demand calculating a lunar time scale and transforming it into other time scales. For users, we present a ready-to-use software package of Lunar Time Ephemeris $\texttt{LTE440}$ that can calculate the Lunar Coordinate Time (TCL) and its relations with the Barycentric Coordinate Time (TCB) and the Barycentric Dynamical Time (TDB). According to the International Astronomical Union Resolutions on relativistic time scales, we numerically calculate the relativistic time-dilation integral in the transformation between TCL and TCB/TDB with the JPL ephemeris DE440 including the gravitational contributions from the Sun, all planets, the main belt asteroids and the Kuiper belt objects, and export data files in the SPICE format. At a conservative estimate, $\texttt{LTE440}$ has an accuracy better than 0.15 ns before 2050 and a numerical precision at the level of 1 ps over its entire time span. The secular drifts between the coordinate times in $\texttt{LTE440}$ are respectively estimated as $\langle \mathrm{d}\,\mathrm{TCL}/\mathrm{d}\,\mathrm{TCB}\rangle=1-1.482\,536\,216\,7\times10^{-8}$ and $\langle \mathrm{d}\,\mathrm{TCL}/\mathrm{d}\,\mathrm{TDB}\rangle=1+6.798\,355\,24\times10^{-10}$. Its most significant periodic variations are an annual term with amplitude of 1.65 ms and a monthly term with amplitude of 126 $μ$s. $\texttt{LTE440}$ might satisfy most of current needs and is publicly available.

astro-ph.IM↗

arXiv2025

Uncertainty-Driven Hierarchical Sampling for Unbalanced Continual Malware Detection with Time-Series Update-Based Retrieval

Android malware detection continues to face persistent challenges stemming from long-term concept drift and class imbalance, as evolving malicious behaviors and shifting usage patterns dynamically reshape feature distributions. Although continual learning (CL) mitigates drift, existing replay-based methods suffer from inherent bias. Specifically, their reliance on classifier uncertainty for sample selection disproportionately prioritizes the dominant benign class, causing overfitting and reduced generalization to evolving malware. To address these limitations, we propose a novel uncertainty-guided CL framework. First, we introduce a hierarchical balanced sampler that employs a dual-phase uncertainty strategy to dynamically balance benign and malicious samples while simultaneously selecting high-information, high-uncertainty instances within each class. This mechanism ensures class equilibrium across both replay and incremental data, thereby enhancing adaptability to emerging threats. Second, we augment the framework with a vector retrieval mechanism that exploits historical malware embeddings to identify evolved variants via similarity-based retrieval, thereby complementing classifier updates. Extensive experiments demonstrate that our framework significantly outperforms state-of-the-art methods under strict low-label conditions (50 labels per phase). It achieves a true positive rate (TPR) of 92.95\% and a mean accuracy (mACC) of 94.26\%, which validates its efficacy for sustainable Android malware detection.

cs.CE↗

arXiv2025

KLIPA: A Knowledge Graph and LLM-Driven QA Framework for IP Analysis

Effectively managing intellectual property is a significant challenge. Traditional methods for patent analysis depend on labor-intensive manual searches and rigid keyword matching. These approaches are often inefficient and struggle to reveal the complex relationships hidden within large patent datasets, hindering strategic decision-making. To overcome these limitations, we introduce KLIPA, a novel framework that leverages a knowledge graph and a large language model (LLM) to significantly advance patent analysis. Our approach integrates three key components: a structured knowledge graph to map explicit relationships between patents, a retrieval-augmented generation(RAG) system to uncover contextual connections, and an intelligent agent that dynamically determines the optimal strategy for resolving user queries. We validated KLIPA on a comprehensive, real-world patent database, where it demonstrated substantial improvements in knowledge extraction, discovery of novel connections, and overall operational efficiency. This combination of technologies enhances retrieval accuracy, reduces reliance on domain experts, and provides a scalable, automated solution for any organization managing intellectual property, including technology corporations and legal firms, allowing them to better navigate the complexities of strategic innovation and competitive intelligence.

cs.IR↗

arXiv2025

Keystroke Detection by Exploiting Unintended RF Emission from Repaired USB Keyboards

Electronic devices and cables inadvertently emit RF emissions as a byproduct of signal processing and/or transmission. Labeled as electromagnetic emanations, they form an EM side-channel for data leakage. Previously, it was believed that such leakage could be contained within a facility since they are weak signals with a short transmission range. However, in the preliminary version of this work [1], we found that the traditional cable repairing process forms a tiny monopole antenna that helps emanations transmit over a long range. Experimentation with three types of cables revealed that emanations from repaired cables remain detectable even at >4 m and can penetrate a 14 cm thick concrete wall. In this extended version, we show that such emanation can be exploited at a long distance for information extraction by detecting keystrokes typed on a repaired USB keyboard. By collecting data for 70 different keystrokes at different distances from the target in 3 diverse environments (open space, a corridor outside an office room, and outside a building) and developing an efficient detection algorithm, ~100% keystroke detection accuracy has been achieved up to 12 m distance, which is the highest reported accuracy at such a long range for USB keyboards in the literature. The effect of two experimental factors, interference and human-body coupling, has been investigated thoroughly. Along with exploring the vulnerability, multi-layer external metal shielding during the repairing process as a possible remedy has been explored. This work exposes a new attack surface caused by hardware modification, its exploitation, and potential countermeasures.

cs.CR↗

arXiv2025

A fully-programmable integrated photonic processor for both domain-specific and general-purpose computing

A variety of complicated computational scenarios have made unprecedented demands on the computing power and energy efficiency of electronic computing systems, including solving intractable nondeterministic polynomial-time (NP)-complete problems and dealing with large-scale artificial intelligence models. Optical computing emerges as a promising paradigm to meet these challenges, whereas current optical computing architectures have limited versatility. Their applications are usually either constrained to a specialized domain or restricted to general-purpose matrix computation. Here, we implement a fully-programmable integrated photonic processor that can be configured to tackle both specific computational problems and general-purpose matrix computation. We achieve complete end-to-end control of the photonic processor by utilizing a self-developed integrated programmable optoelectronic computing platform. For domain-specific computing, our photonic processor can efficiently solve two kinds of NP-complete problems: subset sum problem (far more than 2^N different instances) and exact cover problem. For general-purpose computation, we experimentally demonstrate high-precision optical dot product and further realize accurate image edge detection and MNIST handwritten image classification task with an accuracy of 97%. Our work enhances the versatility and capability of optical computing architecture, paving the way for its practical application in future high-performance and complex computing scenarios.

physics.optics↗

arXiv2025

VSAG: An Optimized Search Framework for Graph-based Approximate Nearest Neighbor Search

Approximate nearest neighbor search (ANNS) is a fundamental problem in vector databases and AI infrastructures. Recent graph-based ANNS algorithms have achieved high search accuracy with practical efficiency. Despite the advancements, these algorithms still face performance bottlenecks in production, due to the random memory access patterns of graph-based search and the high computational overheads of vector distance. In addition, the performance of a graph-based ANNS algorithm is highly sensitive to parameters, while selecting the optimal parameters is cost-prohibitive, e.g., manual tuning requires repeatedly re-building the index. This paper introduces VSAG, an open-source framework that aims to enhance the in production performance of graph-based ANNS algorithms. VSAG has been deployed at scale in the services of Ant Group, and it incorporates three key optimizations: (i) efficient memory access: it reduces L3 cache misses with pre-fetching and cache-friendly vector organization; (ii) automated parameter tuning: it automatically selects performance-optimal parameters without requiring index rebuilding; (iii) efficient distance computation: it leverages modern hardware, scalar quantization, and smartly switches to low-precision representation to dramatically reduce the distance computation costs. We evaluate VSAG on real-world datasets. The experimental results show that VSAG achieves the state-of-the-art performance and provides up to 4x speedup over HNSWlib (an industry-standard library) while ensuring the same accuracy.

cs.DB↗